On 06-Nov-21 08:40, Geoff Huston wrote:
On 5 Nov 2021, at 9:00 pm, Nick Hilliard <nick@xxxxxxxxxx> wrote:
Scott Bradner wrote on 05/11/2021 09:50:
https://www.google.com/intl/en/ipv6/statistics.html - total IPv6 at Google 32.41%
I suspect, without direct evidence to hand, that these figures are heavily weighted by small numbers of larger carriers pulling content from small numbers of larger content networks. There is a long tail of sites and networks that are not ipv6-enabled now, and are likely not to ever deploy the protocol, at no particular disadvantage to themselves.
More generally, I suspect that it's a case of sampling bias. Whenever you generalise from a samped set to a more general observation you need to
understand to what extent the sample is made, and how the nature of the sample impacts the general result. The most striking example in Google’s IPv6 numbers if China, where the 1.34% number reported in that Google URL appears to be anomalously low. Akamai measure some 22% for China [1]. APNIC measures 20% [2] Given the significant user population within China (800M) thats a pretty major difference.
It's also a case of what is being measured. Traffic Volume? Session Count? Eyeballs?
In Google's case, the IPv6 percentage depends only on the IP version number used by a
random sample of Google users:
Lorenzo Colitti, Steinar H. Gunderson, Erik Kline, and Tiziana Refice. 2010. Evaluating
IPv6 adoption in the Internet. In PAM 2010.
http://www.pam2010.ethz.ch/papers/full-length/15.pdf
Brian
All of these measurement programs are capable of providing a decent picture at the general level but each have their failings as you move in closer to get a more granular view.
Geoff
[1] https://www.akamai.com/visualizations/state-of-the-internet-report/ipv6-adoption-visualization
[2] https://stats.labs.apnic.net/ipv6/CN
